1

I am developing an automatic number plate recognition application. I have managed to locate number plate areas by locate rectangular objects in an image. At first I set a minimum height and a width for the number plate of the image so that I could filter the unwanted tiny rectangular bits on the Image. furthermore I use the number plate height and width ratio to filter out false number plates.

but the issue I have is with the values I set for minimum and maximum number plate height and width, because whenever the distance between the camera and the vehicle change the number plate area will not be recognized.

is there a method to overcome this issue?

Mr.Noob
  • 1,005
  • 3
  • 24
  • 58
  • 1
    Its not a simple problem. What do you do about angle and perscpective? Shadows etc ...? Its not just a case of detecting a rectangle with the right aspect ratio. Which, in itself is not trivial. – Jodrell Jul 13 '12 at 12:10
  • 1
    Do you know how much the distance can vary? Geometrically, you need the distance to the plate in order to estimate the real object size (in inches or millimetres, not in pixels). Any way to guess that? – Treb Jul 13 '12 at 12:11
  • 1
    There are a number of license plate standards around the world http://www.worldlicenseplates.com/, which aspect ratios will you be detecting? At what angle to the road will your camera be mounted? What type of lens will your camera use? – Jodrell Jul 13 '12 at 12:16
  • @hypertunes - how do you plan to READ the plate? – Daniel Mošmondor Jul 14 '12 at 17:16
  • Right now I have managed to locate rectangular objects on the image including the number plate. I am thinking of using an OCR engine or may be some neural network. But right now the issue I am having is filtering the false number plate areas – Mr.Noob Jul 14 '12 at 17:22
  • I am asking because OCR engines are notoriously (so they say) unreliable at SCREEN resolution. I'll be glad to know if you find something that can read 'of screen'. – Daniel Mošmondor Jul 14 '12 at 17:38

2 Answers2

4
  1. Identify all lines in the image.
  2. Identify where lines meet. ( = potential corners)
  3. Test 4 and 4 corner combinations for anglesum=360 degrees.
  4. Verify acceptable size ratios.. Are they rectangular and within the plate-range you wanna detect
  5. Extract the potential plate into a new bitmap and correct for prospective. (You do not need 3D calcs, just make a interpolation walk..)
  6. Filter/Adjust
  7. Run OCR
  8. Qualify or disqualify result.
John Conde
  • 217,595
  • 99
  • 455
  • 496
Bjorn Holm
  • 41
  • 2
1

Like is is discussed on the thread the height or width would not be the perfect solution due to the distance between camera and the vehicle. but a rectangular blob detection can be used to overcome this problem to a certain extent.

Mr.Noob
  • 1,005
  • 3
  • 24
  • 58
  • dear i am also doing the same project and i have downloaded a ANPR project developed in C# from this site http://opos.codeplex.com/ but i am not able to run this, i have installed VS2008 and .NET Frame Work less than 4.0, please please help.... – Pir Fahim Shah Jan 23 '13 at 13:46